Statistical Models in Machine Learning for Defense Applications: A Systematic Review and Bibliometric Study

Abstract

The rapid adoption of Machine Learning (ML) in defense-related applications has intensified the need for robust analytical frameworks capable of operating under uncertainty, dynamic conditions, and adversarial environments. Statistical models have long provided foundational principles for inference, prediction, and decision-making, yet their role within modern ML-based defense systems remains fragmented across the literature. This study aims to systematically review and synthesize existing research on the integration of statistical models within Machine Learning approaches for defense-related applications, with a focus on identifying dominant methodologies, application domains, research gaps, and future research directions. A hybrid methodology combining a Systematic Literature Review (SLR) and bibliometric analysis was employed. A total of 130 Scopus-indexed publications were analyzed using Scopus analytical tools and VOSviewer to examine publication trends, collaboration networks, keyword co-occurrence, and thematic clusters. The review followed PRISMA-aligned screening procedures to ensure transparency and reproducibility. The results reveal a steady growth in research output over the last decade, with margin-based and regression-oriented statistical models particularly Support Vector Machines dominating defense-related ML applications, especially in cybersecurity and intrusion detection. Temporal statistical models, such as Markov-based approaches, appear across multiple domains but are often applied in isolation. Explicit probabilistic and uncertainty-aware models remain underrepresented despite their conceptual suitability for military environments. The study identifies a methodological imbalance favoring deterministic classifiers over probabilistic and adaptive statistical–ML frameworks. Future research should prioritize uncertainty-aware, temporally adaptive, and intelligence-driven approaches to enhance the robustness, trustworthiness, and operational relevance of ML-based defense systems.